Using Numerical Optimization for Specifying Individual-Tree Competition Models

In this article we present a method that combines maximum likelihood estimation and nonlinear programming in growth modeling. The method of Hooke and Jeeves is used to discover the optimal specification of a particular competition index type, while statistical software is used to fit the regression model with the given competition index type. The log-likelihood computed by the statistical software is fed back to the optimization algorithm, which alters the specification of the competition index type based on the changes in the log-likelihood. This approach was tested for a mixture of Scots pine (Pinus sylvestris L.) and Norway spruce (Picea abies [L.] Karst.). The characteristics of five different competition index types were optimized. The best model included an index computed from vertical angles formed by a horizontal plane and the tops of competitors. The elevation of the horizontal plane was computed with a species-specific linear regression model using height of the subject tree as the predictor. Pine competitors nearer than 6 m and spruce competitors nearer than 9-10 m were included in the optimal competition index. This study showed that the approach used here is highly efficient.